On 06/07/2018 04:38 PM, Maxime Coquelin wrote:
vhost_vring_call() used rte_mb(), which translates into mfence instruction on x86. This patch changes to use rte_smp_mb(), which changed recently to translate into a locked ADD instruction for performance reason. The measured gain is up to 3% with the testpmd benchmarks. Signed-off-by: Maxime Coquelin <maxime.coque...@redhat.com> --- lib/librte_vhost/vhost.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h index 58c425a5c..1bc99ed3c 100644 --- a/lib/librte_vhost/vhost.h +++ b/lib/librte_vhost/vhost.h @@ -591,7 +591,7 @@ static __rte_always_inline void vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq) { /* Flush used->idx update before we read avail->flags. */ - rte_mb(); + rte_smp_mb();/* Don't kick guest if we don't reach index specified by guest. */if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
Applied to dpdk-next-virtio Thanks, Maxime
